From: Leon Romanovsky <leonro@xxxxxxxxxxxx>
The additions of .doit callbacks posses new access pattern to the
resource entries by some user visible index. Back then, the legacy
DB was implemented as hash because per-index access wasn't needed
and XArray wasn't accepted yet.
Acceptance of XArray together with per-index access requires the refresh
of DB implementation.
Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxxxx>
---
drivers/infiniband/core/nldev.c | 3 +-
drivers/infiniband/core/restrack.c | 101 +++++++++++++++++++++--------
include/rdma/restrack.h | 22 +++++--
3 files changed, 90 insertions(+), 36 deletions(-)
diff --git a/drivers/infiniband/core/nldev.c b/drivers/infiniband/core/nldev.c
index 8ad1116c72de..ff222c12d06d 100644
--- a/drivers/infiniband/core/nldev.c
+++ b/drivers/infiniband/core/nldev.c
@@ -973,6 +973,7 @@ static int res_get_common_dumpit(struct sk_buff *skb,
int start = cb->args[0];
bool has_cap_net_admin;
struct nlmsghdr *nlh;
+ unsigned long id = 0;
u32 index, port = 0;
bool filled = false;
@@ -1023,7 +1024,7 @@ static int res_get_common_dumpit(struct sk_buff *skb,
has_cap_net_admin = netlink_capable(cb->skb, CAP_NET_ADMIN);
down_read(&device->res.rwsem);
- hash_for_each_possible(device->res.hash, res, node, res_type) {
+ xa_for_each(&device->res.xa[res_type], id, res) {
if (idx < start)
goto next;
diff --git a/drivers/infiniband/core/restrack.c b/drivers/infiniband/core/restrack.c
index 46a5c553c624..bf4528b1e940 100644
--- a/drivers/infiniband/core/restrack.c
+++ b/drivers/infiniband/core/restrack.c
@@ -12,6 +12,28 @@
#include "cma_priv.h"
+static int rt_xa_alloc_cyclic(struct xarray *xa, u32 *id, void *entry,
+ u32 *next)
+{
+ int err;
+
+ *id = *next;
+ if (*next == U32_MAX)
+ *id = 0;
+
+ xa_lock(xa);
+ err = __xa_alloc(xa, id, U32_MAX, entry, GFP_KERNEL);
+ if (err && *next != U32_MAX) {
+ *id = 0;
+ err = __xa_alloc(xa, id, *next, entry, GFP_KERNEL);
+ }
+
+ if (!err)
+ *next = *id + 1;
+ xa_unlock(xa);
+ return err;
+}
+
static int fill_res_noop(struct sk_buff *msg,
struct rdma_restrack_entry *entry)
{
@@ -20,6 +42,11 @@ static int fill_res_noop(struct sk_buff *msg,
void rdma_restrack_init(struct rdma_restrack_root *res)
{
+ int i;
+
+ for (i = 0 ; i < RDMA_RESTRACK_MAX; i++)
+ xa_init_flags(&res->xa[i], XA_FLAGS_ALLOC);
+
init_rwsem(&res->rwsem);
res->fill_res_entry = fill_res_noop;
}
@@ -43,33 +70,43 @@ void rdma_restrack_clean(struct rdma_restrack_root *res)
struct rdma_restrack_entry *e;
char buf[TASK_COMM_LEN];
struct ib_device *dev;
+ bool found = false;
const char *owner;
- int bkt;
-
- if (hash_empty(res->hash))
- return;
-
- dev = container_of(res, struct ib_device, res);
- pr_err("restrack: %s", CUT_HERE);
- dev_err(&dev->dev, "BUG: RESTRACK detected leak of resources\n");
- hash_for_each(res->hash, bkt, e, node) {
- if (rdma_is_kernel_res(e)) {
- owner = e->kern_name;
- } else {
- /*
- * There is no need to call get_task_struct here,
- * because we can be here only if there are more
- * get_task_struct() call than put_task_struct().
- */
- get_task_comm(buf, e->task);
- owner = buf;
+ int i;
+
+ for (i = 0 ; i < RDMA_RESTRACK_MAX; i++) {
+ if (!xa_empty(&res->xa[i])) {
+ unsigned long index = 0;
+
+ if (!found) {
+ dev = container_of(res, struct ib_device, res);
+ pr_err("restrack: %s", CUT_HERE);
+ dev_err(&dev->dev, "BUG: RESTRACK detected leak of resources\n");
+ }
+ xa_for_each(&res->xa[i], index, e) {
+ if (rdma_is_kernel_res(e)) {
+ owner = e->kern_name;
+ } else {
+ /*
+ * There is no need to call get_task_struct here,
+ * because we can be here only if there are more
+ * get_task_struct() call than put_task_struct().
+ */
+ get_task_comm(buf, e->task);
+ owner = buf;
+ }
+
+ pr_err("restrack: %s %s object allocated by %s is not freed\n",
+ rdma_is_kernel_res(e) ? "Kernel" :
+ "User",
+ type2str(e->type), owner);
+ }
+ found = true;
}
-
- pr_err("restrack: %s %s object allocated by %s is not freed\n",
- rdma_is_kernel_res(e) ? "Kernel" : "User",
- type2str(e->type), owner);
+ xa_destroy(&res->xa[i]);
}
- pr_err("restrack: %s", CUT_HERE);
+ if (found)
+ pr_err("restrack: %s", CUT_HERE);
}
int rdma_restrack_count(struct rdma_restrack_root *res,
@@ -77,10 +114,11 @@ int rdma_restrack_count(struct rdma_restrack_root *res,
struct pid_namespace *ns)
{
struct rdma_restrack_entry *e;
+ unsigned long index = 0;
u32 cnt = 0;
down_read(&res->rwsem);
- hash_for_each_possible(res->hash, e, node, type) {
+ xa_for_each(&res->xa[type], index, e) {
if (ns == &init_pid_ns ||
(!rdma_is_kernel_res(e) &&
ns == task_active_pid_ns(e->task)))
@@ -157,6 +195,7 @@ EXPORT_SYMBOL(rdma_restrack_set_task);
static void rdma_restrack_add(struct rdma_restrack_entry *res)
{
struct ib_device *dev = res_to_dev(res);
+ int ret;
if (!dev)
return;
@@ -174,10 +213,13 @@ static void rdma_restrack_add(struct rdma_restrack_entry *res)
kref_init(&res->kref);
init_completion(&res->comp);
- res->valid = true;
down_write(&dev->res.rwsem);
- hash_add(dev->res.hash, &res->node, res->type);
+ ret = rt_xa_alloc_cyclic(&dev->res.xa[res->type], &res->id, res,
+ &dev->res.next_id[res->type]);
+
+ if (!ret)
+ res->valid = true;
up_write(&dev->res.rwsem);
}
@@ -234,12 +276,15 @@ void rdma_restrack_del(struct rdma_restrack_entry *res)
if (!dev)
return;
+ if (!xa_load(&dev->res.xa[res->type], res->id))
+ goto out;
+
rdma_restrack_put(res);
wait_for_completion(&res->comp);
down_write(&dev->res.rwsem);
- hash_del(&res->node);
+ xa_erase(&dev->res.xa[res->type], res->id);
res->valid = false;
up_write(&dev->res.rwsem);
diff --git a/include/rdma/restrack.h b/include/rdma/restrack.h
index 8f179be9d9a9..b0720d54cd7e 100644
--- a/include/rdma/restrack.h
+++ b/include/rdma/restrack.h
@@ -13,6 +13,7 @@
#include <linux/completion.h>
#include <linux/sched/task.h>
#include <uapi/rdma/rdma_netlink.h>
+#include <linux/xarray.h>
/**
* enum rdma_restrack_type - HW objects to track
@@ -48,7 +49,6 @@ enum rdma_restrack_type {
RDMA_RESTRACK_MAX
};
-#define RDMA_RESTRACK_HASH_BITS 8
struct rdma_restrack_entry;
/**
@@ -61,9 +61,13 @@ struct rdma_restrack_root {
*/
struct rw_semaphore rwsem;
/**
- * @hash: global database for all resources per-device
+ * @xa: Array of XArray structures to hold restrack entries.
+ * We want to use array of XArrays because insertion is type
+ * dependent. For types with xisiting unique ID (like QPN),
+ * we will insert to that unique index. For other types,
+ * we insert based on pointers and auto-allocate unique index.
*/
- DECLARE_HASHTABLE(hash, RDMA_RESTRACK_HASH_BITS);
+ struct xarray xa[RDMA_RESTRACK_MAX];
/**
* @fill_res_entry: driver-specific fill function
*
@@ -71,6 +75,10 @@ struct rdma_restrack_root {
*/
int (*fill_res_entry)(struct sk_buff *msg,
struct rdma_restrack_entry *entry);
+ /**
+ * #next_id: Next ID to support cyclic allocation
+ */
+ u32 next_id[RDMA_RESTRACK_MAX];
};
/**
@@ -108,10 +116,6 @@ struct rdma_restrack_entry {
* @kern_name: name of owner for the kernel created entities.
*/
const char *kern_name;
- /**
- * @node: hash table entry
- */
- struct hlist_node node;
/**
* @type: various objects in restrack database
*/
@@ -120,6 +124,10 @@ struct rdma_restrack_entry {
* @user: user resource
*/
bool user;
+ /**
+ * @id: ID to expose to users
+ */
+ u32 id;
};
/**
--
2.19.1
